A typical commercial search engine may employ a model that is produced in a machine-learning process. The machine-learning process may train the model on the basis of click-through data. The click-through data, in turn, identifies queries submitted by users, together with selections (e.g., “clicks”) made by the users in response to the queries.